2.2.3 Differential
Coexpression Genes and
Gene Pair Identification
Complementary to traditional differential expression analysis, the
differential coexpression genes and gene pairs can further help to
explain the underlying molecular mechanisms for a biological prob-
lem from the systematic level. Differential coexpression genes are
defined as genes whose correlated expression pattern differs
between classes. Table3 gave the widely used algorithms adopted
by DCG and DCL identification tools. Yu et al. [12] reported that
the traditional differential expression analysis just identified a part
of T2D-associated genes, and a considerable amount of genes were
identified by differential coexpression analysis as DCGs. Besides the
DCGs, the DCLs can also be identified and validated to be very
important. For example, RNA levels of prostate cancer biomarker
geneAMACRhave found to have positive with tumor suppressor
genePTEN in normal tissue but not in prostate cancer tissue
samples [22]. So many differential coexpression genes and gene
pair identification tools have been made based on different algo-
rithms [8, 12, 16, 22–29]. Among them, DCGL is a very com-
monly used tool to identify differential coexpression genes and
gene pairs simultaneously. And there are two methods of DCp
and DCe to identify DCGs in which DCp used the length-
normalized Euclidean distance to measure the difference of gene
interaction with its neighbors and then calculate the significantp-
value using a permutation test, while the method DCe adopted the
hypermetric model to test whether the test gene enriched signifi-
cantly more differentially coexpressed gene pairs.
Lai et al. extended the traditionalF-statistic to ECF-statistics to
identify differential gene-gene coexpression pattern. Choi et al. and
Yoon et al. adoptedz-score and Fisher’s z-transformed score to
measure the difference of gene pairs under two states and then
applied methods to cancer research.
DiffCorr calculates correlations in each condition and uses the
difference in z-transformed correlation coefficients to calculatep-
values. EBcoexpress uses an empirical Bayesian approach and a
nested expectation-maximization algorithm to estimate the poste-
rior probability of differential correlation between gene pairs. Dis-
cordant fits a mixture distribution of correlation classes in eachTable 3
Methods for differential coexpression genes and gene pair identification
Methods Identify genes Identify gene pairs References
Z-score No Yes [22, 23, 26, 27]
F-statistics No Yes [22]
Euclidean distance and hypergeometric model Yes Yes [8, 12, 16]
Nested expectation-maximization algorithm No Yes [28, 29]162 Bao-Hong Liu